Daily Horoscope: Polaris Software- A Critical Analysis.
Note From George: Many of you already know that for the last several days I have been wading into the astrology debate over at Jason’s blog, Lousy Canuck.
During the original debate, we were invited by Jamie Funk to join in on the discussion at his astrology blog where I first encountered James Alexander and his brief explanation of Polaris; a computer software developed for “rectification”. James has now joined the debate on Jason’s blog, and has once again brought up Polaris as evidence of astrology’s ability to make falsifiable predictions. During this debate, I have offered James the opportunity to test Polaris as a proof of astrology and he seems genuinely interested in putting it to task. In the interests of giving a fair shake to James, I would like to give him the opportunity to guest post his own interpretation of Polaris, which I will not edit save a disclaimer that the views are his, and post it on my site. I welcome his comments about my interpretation, which I am offering here. I would forewarn readers that this is a 4000 word post with no jokes and little pointed language, and will likely be a tl;dr for anyone not interested by astrology or with vested interest in our ongoing discussion. Feel free to read my Summary just above the fold to get a brief overview of this post. Unless you are James. Then you should read it and explain in some detail what parts are factually wrong, as well as proofread it for spelling and grammar. (that’s my only joke folks, you have been warned) All quotes or information attributed to James is available at the Polaris link or in comments on my blog and Lousy Canuck. I will be happy to clarify the source upon request.
Jason has offered to post on his blog the parameters and eventual results of this test of Polaris on his site, once James and I have agreed on terms and begun the test.
- Rectification can cause “warm readings” as opposed to “cold readings”, the potential for non-astrologically gained information and/or the discounting of information should be considered as fostering confirmation bias.
- It is feasible to create a PRNG (Pseudo-Random Number Generator) that would perform better than chance without the aid of astrology.
- The odds quoted by James are fundamentally flawed
- Many of his corollary statements are misdirecting, flawed or incorrect
- By widening the scope of what would pass for a “hit” for Polaris, the odds of the “uncanny” become far better.
- Polaris is deserving of a test in spite of my basic criticisms
Polaris Software:Rectification, Probabilities, and The Nature Of Not Being Random- A Critical Analysis.
What is Polaris? What is Rectification?
Polaris is a rectification software developed by Isaac Starkman, using the methods of the late astrologer and researcher Alexander Marr. It seeks to use a logarithmic equation to weigh potential birth times as being more or less likely based on a point scale, using supplied events of importance from the subject’s life.
Some astrologers believe rectification to be an important process in giving clients a proper astrological reading. The belief is that many people are unsure of the exact time of birth and therefor their chart will be skewed by insufficient or improper data. Rectification seeks to correct this error by calculating a probable “real” birth time by weighing the probability of past known events against a list of possible birth times using astrological methods.
Rectification, in relation to Polaris, is a process where a person’s birth time is calculated (rectified) by a computer program with the aid of additional important dates in that person’s life. The subject need only give their birth date, hopefully a rough idea of the time of birth, and as many dates of importance from their life as possible and the program goes to work calculating the “real” birth time, even to the second.
James has offered us a list of some of the events in your life the program considers.
Below is a list of the types (categories, if you will) of events that can help in the rectification process with Polaris:
1 Birth of Brother
2 Birth of Sister
3 Birth of Son
4 Birth of Daughter
5 Birth of Grandson
6 Birth of Granddaughter
7 Marriage/Engagement/Love-affair (for a M)
8 Marriage/Engagement/Love-affair (for a F)
9 Marriage of Son/Daughter
11 Death of Father/Grandfather
12 Death of Mother/Grandmother
13 Death of Son
14 Death of Daughter
15 Death of Wife/Friend
16 Death of Husband/Friend
17 Death of Brother
18 Death of Sister
25 Travel Overseas (positive)
26 Travel (positive)
27 Travel (negative)
28 Mobilization (Military)
29 Demobilization/Release (Leave Military/Prison)
36 Gambling Losses
37 Gambling Gains
40 Promotion Army
Confirmation Bias And The Need To Be Critical
James admits that many people do not have an exact birth time, or that the real birth time is rounded to the nearest fifteen minute interval and this is part of the need for good rectification software. I have given an example of this software in action (which I pulled directly from an article authored by James) in my last post. I will be using it as an example again later in this post.
I do have some logical concerns with the claims made by James in regards to Polaris; chief among them is the issue of confirmation bias. I hope it is safe to assume that the vast majority of people seeking a rectification already have astrological leanings. Based on this fact, they are already primed to accept a different birth time than the one that is on their birth certificate. This is certainly true for James’ case study in the article he posted that I have linked to above. In this case, the “very delightful lady” was given a birth time that was “over 20 minutes difference from the time given on the birth certificate” (bold by the original author).
James has already conceded that many people do not even have an inclining of their birth time. For these people, any time Polaris calculates must seem correct. Even if we take James’ estimate that most people’s birth times are rounded by hospital staff to the nearest fifteen minute interval, then the possibilities for “hits” on recorded birth times greatly increases. So the credulity of the person having a rectification is certainly going to be a factor in whether or not they accept the software over documentation, where available.
There is also a concern that the information given for the rectification would surely give an astrologer more information with which to create an uncanny reading that will only serve to amplify the confirmation bias. Having a wealth of additional information, an astrologer may be willing to discount things in a reading that do not correlate with these newly learned events. Conversely, the astrologer might be inclined to add information or interpret the information differently in light of these known facts. In effect, the client has turned what might be a “cold reading” into a “warm reading” by supplying the astrologer with additional information. As I have said elsewhere, it would even be degrees easier to extrapolate predictions of future events given a talented “people-reader” and a known history from which to draw information. My fear is that the astrological reading then becomes an exercise in astute observation of a person and not of their charts.
In order to scrutinize the software then, we must subject it to a test where confirmation bias is not present.
James has indicated that it is the shear uncanny nature of the odds involved that has convinced him of the efficacy of Polaris and the truth of astrology. If the information James has given me is even partially correct, Polaris does appear to have some merit for study. Chief among James’ claims are that Polaris is consistently exact or within a negligible margin, that Polaris is consistently beating odds of 1:1440, and that these observations cannot be explained away by a “random time generator”.
Not Random, Not Exact, And Not Equal
Polaris is not a “random time generator”. At no time during my conversation with him have I or anyone else said it was. Polaris, at worst, contains a pseudo-random number generator; given the same information it will always arrive at the same time. Given the same important dates but two different constraints to weigh potential birth times, Polaris should also give the same “peak value” to numbers contained in both sets. There is no magic in a birth time that is present in both sets floating to the top of both lists.
I would argue that given enough data a programmer could create a PRNG that performs better than chance at arriving at birth times without the aid of astrology. James’ claim that the odds in a 24 search for a potential birth time is always 1:1440 is likely not accurate. The fact that times might be negligibly close to the recorded time and still considered a hit changes those odds drastically. If we expand the window to just three minutes either side of a recorded birth time, the odds become approximately 1:206. James has implied in comments that he would consider an inexact match within a small margin to be a “hit” for his software. Depending on the margin we can agree is negligible, we will soon see that odds are always better than 1:1440.
Yet even this calculation takes at face value the fact that each potential birth time has an equal probability. Given that we are dealing with a process of the human body, it is reasonable to believe that births are more common during certain hours in a day. Human rhythms and schedules would surely make some birth times more common than others. This change in probability means that each of the 1440 minutes in consideration does not have an equal value as a potential birth time. Now combine this with another piece of information offered by James. In a comment, James said that it is quite common for hospitals to round birth times to the nearest 5 or 10 minute interval, and he offered as proof a database that shows that those times are much more common than the times between these numbers. An astutely designed PNRG then could take advantage of this information to always perform better than chance.
Given these facts as a prism with which to view calculated odds, let’s now take a closer look at some of James’ specific examples. James is quite proficient at math I’m sure, he admits to even winning a “1st place Illinois State Mathematics Award”. I am impressed by his ability to use numbers, I just believe he is stacking the numbers in his favor. I would like to discuss a few times where James has quoted probabilities as well as dissect the probabilities of the few documented examples he has given. For the sake of offering comparable odds, in each of the following examples I will assume that each potential birth time has an equal probability even if this is no the case.
By The Numbers: Dissecting The Examples Given By James
James gives us an example of his program at work in the article he wrote outlining Polaris software. I have already briefly discussed this in my last post, which James has said is wrong, so it bears some repeating and analysis of how I arrive at my numbers:
A very delightful lady had a birth certificate time (8:15 PM CWT), but was skeptical as to its accuracy as it appeared rounded off. Since her Ascendant changed signs close to that time, she wanted to be sure. Using the birth certificate time, she had Sagittarius 29° 32’ rising. A birth time two minutes later would have given Capricorn rising.
She sent me over 40 events from her life, mostly with exact dates. I took 38 of these events (the ones where the dates were most accurately known) and entered each of them into Polaris. I gave the software a search range of an hour on either side of the supposed birthtime. This entering of events goes quite quickly. In about a minute (time dependent on computer speed), Polaris examined every 8 seconds in birthtime throughout the range and gave the following table:
Possible birth times are sorted by the weighted score (as shown here), with the highest score on this particular list, 994, indicating the most likely time and the lowest score, 775 indicating the least likely. In this example, we can see that 0:52:52 UT is the most likely birthtime, over 20 minutes difference from the time given on the birth certificate, and clearly indicating Sagittarius rising as correct!
The letters A-E are breakdowns of Polaris’ findings. We can see that the indicated time is the highest value, not only in T, but also in A, B, C, and E. We must cross-check ANY findings through Polaris with other dynamic systems. Checking in one system is never enough.
In this example, James clearly documents his parameters used as possible birth times. He says that he used 8 second intervals over a period of an hour on either side of her recorded birth time. Given that this client received a birth time roughly 22 minutes divergent from her birth time, what are the odds that this would happen?
-There are 900 choices available to the program (7.5 per min x 120 minutes= 900)
-There are 330 choices available during a 22 minute span either side (44 x 7.5=330)
– So if this is subject to the same “odds” James applies to more uncanny examples the odds rest at just a bit better than 1 in 3.
My previous dissection of this example showed that for someone who may find a ten minute window from the recorded birth time uncanny the odds were 1 in 6. But what if it had of gotten the birth time to within one minute, or even creepier, to the exact minute?
The odds in this case are 1 in 40 that we should arrive at a number within just one minute divergent and 1 in 120 that we should arrive at the exact time. These are impressive odds, but certainly seem like better odds than James likes to cite.
To be fair, James never cites odds for this example, except in a roundabout way at the end. To quote him:
By the way, if we had run this rectification without having any idea of the time (that is, with a search range of the whole birth day), it gives the correct time as the 2nd entry, which is impressive considering Polaris examined 10,800 moments in time.
This is mildly misdirecting the reader. The odds are still stacked against the program, but the program only really had to examine 9,900 moments in time, remembering that it’s logarithm had already discounted 899 other numbers as having less probability than the first choice. If Polaris’ agreed birth time is rated number 2 after this second assessment, then it should also count as a potential strike against the program for someone with absolutely no idea of their recorded birth time; the best weighted time is not necessarily the right one. Odds that Polaris should rank the first rectified time as #2 on a second 24 hour test: 1in 4950….if this were a random number generator. The trick here is that the second time around you have already found one number in the set with a relatively high peak value. The odds are actually greater that this number should not end up quite high on the list. If you cast you attention to the chart provided by Mr. Alexander, you see that the “winning” birth time has a “peak value” of 994, while the second highest choice has a value of 862. If there is a 132 point difference between choice #1 and choice #2 and you plug the numbers back into an expanded set in the same logarithm, what do you want to bet that the original choice will float relatively near the top of the pile? So our “odds” in this case are heavily skewed in favor of the time that has a very high peak value.
The actual odds escape my ability to derive a probability, but I can safely assume that it is far better than 1 in 25.
In all fairness to James, I chose an example where I knew the odds were far better than he would let on, and the subject of the rectification in this case was willing to accept a birth time that was relatively divergent from the recorded one. So what about a situation where Polaris got the time exactly to the certificate? Surely the odds must be unbelievably stacked against this.
To state the question succinctly in James’ own words:
You still have not explained HOW the program can EVER give the correct birthtime, based on events from someone’s life. (Remember that chance levels say that this should happen ONCE in every 1,440 attempts.)
From an anecdote that James provided on the astrology thread:
A very nice lady gave me eleven events, including births of two brothers, deaths of grandparents, an illness… She told me that she was born (according to her Mother) between 10:30 and 11 am. I did the rectification and it is normal in these circumstances, to expand the “believed” timeframe, to make sure that the actual birthtime isn’t just outside the perceived range and thereby missed. The time that I came up with (ie. Polaris), 10:26:14, was before the range that she had given. When she came for our discussion about her life, she apologized that she had given me the wrong time range and said, “I found my paperwork, I was born 4 minutes before the time range I gave you.”
This was not a search out of 24 hours, because she had a believed timerange, which I slightly expanded in order to be sure the actual time was included.
In the hour and a half search, the correct birthtime was the highest “peak” on the list. In a 24 hour search that I just ran with the same data, the correct birthtime is number 2. Considering that the program looked at 10,800 moments in time, this is phenomenal performance. (BOLD is mine)
Disregarding the obvious similarities between the previous story and this one, there are some major differences. Polaris gave an exact birth time to match the paperwork. But are the odds really 1 in 1440?
To calculate this, I am going to assume that James went by 8 second intervals just as he did in the first example, and just as he does when he “re-plugs” the “winning time” into a 24 hour cycle in the last paragraph of this example.
With an hour and a half search frame he mentions, there are 675 possible birth times to consider.(7.5 x 90)
For Polaris to stumble across the exact time would be odds of 1 in 90 (7 or 8 numbers in that exact minute out of 675 possible choices).
So our odds here, where someone eventually did verify Polaris’ findings, are 1:90. So I believe I have explained how it is possible for the odds to be better than 1 in 1440. I would argue that the astrologer’s confirmation bias in this case led them to imagine greater odds than were actually against a “hit”. I also postulate that the other 4 times James has had this happen, the odds were far better than he imagines. They are still phenomenal odds. I just question why one has to puff up their odds to make their case.
As an aside, the same criticism I made earlier of the second 24 hour test is the same here. If you already have a number with a high peak value, and you use the same logarithm a second time, that number should still rise to very near the top of the list. I would be more impressed if it did not, so this is hardly proof of “phenomenal performance”.
I am assuming from James’ penchant for inflating statistics that the previous example counts as one of his 5 direct hits for Polaris. Even if this is the only example of the five where it arrived at the right time with odds better than 1:1440, it drastically changes the odds he states of 1: 6,191,736,400,000,000. If every other time this happened, he arrived at the answer with 1:1440 odds, then the actual probability is 1:386,983,521,000,000. That means he is off by a hell of a lot. I am not, however, so confident that the odds were always 1:1440 given his track record with statistics, and if we assume that 3 of the 5 were 1:90 odds and 2 were 1:1440 odds then the actual probability is 1:1,511,654,400,000. Now we are talking about puffing up odds by degrees. What if all the 5 were arrived at by the method he used in example 1 and 2? Let’s for the sake of his pride say that the average odds were more like 1:120, like the first example. Now the odds are 1:24,883,200,000.
You can see how changing the odds to values he has evidenced are more likely can really change the probability, even if the new probabilities are still astronomical. Are they really astronomical though? James has said that he has performed 100 or so rectifications. If we discount just half of these and assume no birth time was available, his program would now be arriving at the right time only once out of every 10 tries. At 1:120 odds that the program will arrive at the right time any one time, I calculate the odds that it would be correct to the minute just 1:248,832.
That is a calculation for 5 direct “to the minute” hits out of 50 tries with a search range of an hour either side for each try, assuming again that each potential time in the cycle has an equal probability.
George, you say, what about his “crown jewel”, the rectification that was performed on Ken Haining? Honestly, I don’t have an easy answer for that one. It doesn’t surprise me that this is also the example that James offers the least specifics of his methods. Only that the odds were 1:1440. I am forced to assume the best, that he was looking for a number to the minute in a 24 hour cycle from the start. If he was not, if he chose that number out of a 2 hour or 1 ½ hour cycle, then plugged it into a 24 hour cycle to test it; I will leave it to the reader to interpret the true odds. Why Polaris makes the “correct time” #2 in our previous 2 examples of a 24 hour search and #1 in this one is also a mystery.
He also talks about the odds of splitting the events into two groups and separately arriving at the same time. It is again quite impressive, especially if he is working in a 24 hour cycle. I can’t say for sure that he is. In a situation where two different 6 event sets each arrive at the same birth time as the full 12 considered together, I don’t think the odds are again as improbable as he lets on. I can’t personally prove this statement with my limited math skills, but I can imagine that each event is given a value at each moment in the cycle that the program is asked to consider; and this number is fixed by the logarithm built into Polaris. Given this, the probability is better than chance that two groups of six will arrive at the same number because the logarithm is not randomly assigning values to each event. The assumption here is that each event should have a higher than average value at the rectified birth time. A group of any six of these events should then have a relatively good probability of arriving at the same time, and a very good probability of having perhaps 2 or 3 times in common in any list of the 5 best choices.
So Where Does This Leave Polaris?
I do not wish to dismiss the apparent slim odds that his program seems to overcome. However, I have given examples of how these apparent “astronomical odds” might be far less impressive than any of us imagine. I have also shown that it might be wrong to assume that any given rectified time has an equal probability. I would consider 1:90 odds to be long odds in their own right, yet if we combine that calculation with the greater likelihood of certain times it further reduces the “uncanny” nature of arriving at a negligibly different time than the recorded one.
I take issue, however, with his gross overestimate of the odds against his program. There is no reason short of being ignorant or purposefully misdirecting to change the probabilities in this way. The only time Polaris could face odds of 1:1440 would be in a 24 hour search with each potential time afforded equal possibility. I am sure this is not the case, and I am sure that James does not believe this to be the case either-even if for very different reasons. He has offered more than one example where he did not plug his information into a 24 hour cycle, and I believe it is logical to conclude that this is the rule rather than the exception. If his anecdotal evidence is true, however, then most certainly Polaris shows promise as a potential positive evidence for astronomy.
I am sure James will find much to take issue with in this summary, yet I believe I have given an answer to many of his pointed questions. I have explained how, given the information James has himself provided us, the odds of arriving at an exact birth time might be better than 1:1440. I have explained how his calculations of odds on more than one occasion appear to be misleading. Most pointedly, I have shown that he seems just as content with a time that is 22 minutes divergent from the recorded time as he is with an exact match.
James never explains to us why his “direct hits”, those 5 times that the program got it just right, should have so much weight given that he himself claims that there “are many, many times that are rounded off, estimated, and improperly recorded.” It would stand to reason that those 5 times have an equal likelihood of being correct on paper compared to, let’s say, the “very delightful lady” who was 22 minutes divergent from Example 1. He never tells us how many of the 100 or so customers he has performed rectification on had documentation that contradicted Polaris.
There is likewise no explanation as to why, when he has offered specifics and my calculations have shown his odds to be inflated, the second 24 hour calculation has resulted in a 2nd place ranking for his assumed “real” birth time in light of his assertion that a strict 24 hour calculation for Ken results in a 1st place ranking for his known birth time.
In light of these considerations, I find James’ claim that Polaris is a useful and precise tool to be lacking, yet the uncanny nature of some of his anecdotes shows some promise. I hope that we might be able to address these obvious shortcomings in our ongoing conversation as well as during the testing period.